def predict_validation(self, predictor, tag):
valid_pred_result = common.predict_proba_or_label(predictor, self.Xvalid, self.yvalid.index, tag)
valid_pred_result.to_csv("meta_features/{}_validate.csv".format(tag), index_label="id")
colname_proba = "{}_proba".format(tag)
if colname_proba in valid_pred_result:
yvalid_pred_proba = valid_pred_result.loc[:, colname_proba]
valid_auc = roc_auc_score(self.yvalid, yvalid_pred_proba)
valid_accuracy = accuracy_score(self.yvalid, yvalid_pred_proba > 0.5)
else:
assert valid_pred_result.shape[1] == 1, "only one label column"
yvalid_pred_label = valid_pred_result.iloc[:, 0]
valid_accuracy = accuracy_score(self.yvalid, yvalid_pred_label)
valid_auc = np.NaN
self.stats_file.log(tag, valid_accuracy, valid_auc, str(predictor))
def run_once(self, predictor, tag):
# ---------- fit on all training dataset to get the model
predictor.fit(self.Xtrain, self.ytrain)
common.simple_dump("meta_features/{}.pkl".format(tag), predictor)
print "\tModel[{}] learnt and saved".format(tag)
# ---------- predict on validation
self.predict_validation(predictor, tag)
print "\tModel[{}] predicted on validation".format(tag)
# ---------- predict on test
test_pred_result = common.predict_proba_or_label(predictor, self.Xtest, self.ytest.index, tag)
test_pred_result.to_csv("meta_features/{}_test.csv".format(tag), index_label="id")
print "\tModel[{}] predicted on test".format(tag)
# ---------- generate meta features
metafeatures = crossval_predict(predictor, self.Xtrain, self.ytrain, tag)
metafeatures.to_csv("meta_features/{}_train.csv".format(tag), index_label="id")
print "\tMeta-features generated from Model[{}]".format(tag)
def predict_validation(self, predictor, tag):
valid_pred_result = common.predict_proba_or_label(
predictor, self.Xvalid, self.yvalid.index, tag)
valid_pred_result.to_csv("meta_features/{}_validate.csv".format(tag),
index_label='id')
colname_proba = '{}_proba'.format(tag)
if colname_proba in valid_pred_result:
yvalid_pred_proba = valid_pred_result.loc[:, colname_proba]
valid_auc = roc_auc_score(self.yvalid, yvalid_pred_proba)
valid_accuracy = accuracy_score(self.yvalid,
yvalid_pred_proba > 0.5)
else:
assert valid_pred_result.shape[1] == 1, 'only one label column'
yvalid_pred_label = valid_pred_result.iloc[:, 0]
valid_accuracy = accuracy_score(self.yvalid, yvalid_pred_label)
valid_auc = np.NaN
self.stats_file.log(tag, valid_accuracy, valid_auc, str(predictor))
def run_once(self, predictor, tag):
# ---------- fit on all training dataset to get the model
predictor.fit(self.Xtrain, self.ytrain)
common.simple_dump("meta_features/{}.pkl".format(tag), predictor)
print "\tModel[{}] learnt and saved".format(tag)
# ---------- predict on validation
self.predict_validation(predictor, tag)
print "\tModel[{}] predicted on validation".format(tag)
# ---------- predict on test
test_pred_result = common.predict_proba_or_label(
predictor, self.Xtest, self.ytest.index, tag)
test_pred_result.to_csv("meta_features/{}_test.csv".format(tag),
index_label='id')
print "\tModel[{}] predicted on test".format(tag)
# ---------- generate meta features
metafeatures = crossval_predict(predictor, self.Xtrain, self.ytrain,
tag)
metafeatures.to_csv("meta_features/{}_train.csv".format(tag),
index_label='id')
print "\tMeta-features generated from Model[{}]".format(tag)